The Andrej Karpathy interview 0:00:00 – AGI is still a decade away 0:30:33 – LLM cognitive deficits 0:40:53 – RL is terrible 0:50:26 – How do humans learn? 1:07:13 – AGI will blend into 2% GDP growth 1:18:24 – ASI 1:33:38 – Evolution of intelligence & culture 1:43:43 - Why self

This is giving strong Bayesian estimation vibes 😂

Here's a look at the Most Nominated Games at #TheGameAwards Clair Obscur: Expedition 33 is the most nominated game in TGA history with 12 nominations.

Diffusion models (for now) only generate bitmaps — what are you even panicking about? 4K is still just a bitmap 😂

The Ilya Sutskever episode 0:00:00 – Explaining model jaggedness 0:09:39 - Emotions and value functions 0:18:49 – What are we scaling? 0:25:13 – Why humans generalize better than models 0:35:45 – Straight-shotting superintelligence 0:46:47 – SSI’s model will learn from deployment

😆

If we’re really “all-in on AGI,” stop building products and start rebuilding workflows. Product-based thinking blocks you; workflow-based design actually scales your features automatically with model upgrades. 💪

A diffusion model is a wandering artist with unstable vibes. Google keeps trying to domesticate it inside Office apps. The charm of banana-flavored slide components will die quickly, and generative PPT will become yet another dead end for diffusion models next to video.

Google DeepMind Gemini 3 Pro 黑客松，每人将获得价值 10,000 美元的 Gemini API 积分（不是现金，但也值钱）。 即日起至 12月12日（只有一周时间冲刺！） 开发平台： Google AI Studio（活动期间免费使用 Gemini 3 Pro 构建） 选出 50 位 优胜者，每人奖励 $10,000 美元 (约 7.2w 人民币) 的

高市政権が増税に前向き姿勢を示さない限りは円安・債券安が続くでしょ、という趣旨のリポートが到着。

Turns out ‘vibe coding’ is great for shipping apps — just not for inventing new algorithms.

It misses every crucial detail, improvises random code, and then hides everything behind a catch-all fallback just to make the tests pass.

Clair Obscur: Expedition 33 has won Game of the Year at #TheGameAwards! bit.ly/3MBRTKX

夜渡りの皆様は ナイトレインやってた時間を勉強に捧げてたら どの資格取れそうでした？！ 僕は臨床工学技士ってのいけたかも知れませんね！

score matching 学习的是真实的图片，而不是图片中的真实。transformer 预测下一句话也是同理：学习的是真实语言，而不是语言中的真实。只是语言的抽象层级更高，又是思想的边界，人们更容易混同语言与真实本身。但值得兴奋的是，“真实感”确实作为一种规律，存在于像素之中，并且可以被学习。

“真实感”可能只是一个“流形”，每个人都有自己的流形，人们又共享某些流形，而这些流形是光滑、连续、可操作的。关键是这些流形是怎么来的。

啊我不是很喜欢“流形”这个概念的语义的演绎，就让它留在概率论中吧。

“世界并不是光滑的而是只有光滑的部分能被机器学习到”，这样感觉好多了😂

“AGI”不是大杀器，甚至不是效率提升工具。他是给人们初尝“权力（知识平权)”机会的众筹App。

you can just tweet “Claude code” and immediately get 1000 likes